class JoelNguemeta:
role = "AI Platform Engineer"
focus = ["LLM Serving", "MLOps", "Distributed Systems", "Kubernetes"]
building = "Healthcare AI platforms running on-premise LLMs in production"
stack = {"backend": ["Django", "FastAPI", "Spring Boot"],
"ml": ["MLflow", "QLoRA", "RAG", "vLLM"],
"infra": ["Kubernetes", "ArgoCD", "Terraform", "Helm"]}
principle = "Distributed systems fail in distributed ways — design for it from day one."
def current_mission(self) -> str:
return "Shipping AI infrastructure so solid, teams forget it exists."|
Plateforme médicale IA — en production dans des hôpitaux réels 🇨🇲🇧🇫 LLMs médicaux servis on-premise (Meditron-7B, BioMistral-7B), pipeline RAG (ChromaDB + LlamaIndex), anonymisation PII (Presidio), fine-tuning QLoRA, et fallback déterministe sur le protocole de triage Manchester.
|
SaaS de notifications multi-tenant Modèle d'isolation hybride (shared-schema → dedicated-schema par tier), envelope encryption des credentials, lifecycle JobExecution complet,
|
|
RBAC + ABAC + multi-tenant, grade entreprise
|
Infra de training & serving multi-cluster Clusters locaux kind/k3d pour workflows MLOps, GitOps avec ArgoCD, tracking d'expériences et model registry MLflow, observabilité Prometheus/Grafana.
|
- 📄 Multi-cluster Kubernetes patterns for ML workloads
- 📄 Serving medical LLMs on-premise: lessons from production
- 📄 Django RBAC + ABAC: a practical guide to multi-tenant auth
Observable · Fault-tolerant · Reproducible
L'objectif est toujours le même : une infrastructure si solide que l'équipe arrête d'y penser et se concentre sur le produit.



